On October 1, Anthropic shipped Claude Code v2.1.287 with a feature that changes what the tool is: Mods. A mod is a small JavaScript or TypeScript function that lives inside a plugin and hooks directly into Claude Code's internal events — the prompts, tool calls, permission requests, and interface rendering that make up the agent loop.
The practical difference from the older settings hooks is depth. A hook could run a shell command when something happened. A mod runs inside the event itself: it can act before an event, after it, instead of it, or wrap it on both sides. That means a mod can rewrite a prompt before it reaches the model, block or retry a tool call, approve or deny a permission request, redact secrets from tool output, and even draw or replace parts of the UI while a session is running. When several mods hook the same event, they run as a middleware chain in load order — the first mod to load sees the event first and the result last. Three built-in features — the diff pane, the agents.md , and telemetry — were converted to mods at launch, so /diff is now a mod you can disable or replace like any third-party extension. The release also ships a new built-in mod, "You should know": a side agent that watches the session and flags things you or Claude might miss. It is off by default and turns on with a /plugin command. On Team and Enterprise plans, a built-in sec-default mod loads first to block risky moves like overriding a permission denial. You install mods through the /plugin command in the CLI or the desktop app, and you can write one yourself — or ask Claude Code to write it for you.
Here is what actually matters if you use Claude Code against a Salesforce org.
1. A mod can redact org secrets before Claude ever sees them.
When Claude runs cat .env or reads a config file, the output lands in the conversation — keys, tokens, and session strings included. A mod can sit around every Bash and Read call, let the tool run, and replace anything that looks like a secret with [REDACTED] before the model reads the result. For a Salesforce team, this is the difference between "don't paste your org password in chat" and a rule the tool enforces itself. If your team is already set up to run Claude Code against sandboxes, this is the next control to add — here's the setup guide we published if you still need the terminal, desktop app, or VS Code walkthrough.
2. A mod can be a deployment guardrail.
Anthropic's examples include requiring confirmation before commands alter production configuration and recording tool activity for audit. Translate that to Salesforce: a mod that denies sf project deploy start whenever the target org alias is your production alias, or one that forces a second confirmation before any command that writes to an org. That is a stronger promise than a CLAUDE.md line asking Claude to "never target production" — it is enforced at the event level, where a prompt rewrite can't be talked past.
3. A mod can inject your org's coding conventions.
Instead of repeating "use Custom Metadata, never hardcode record IDs; bulkify every trigger" in every prompt, a mod can rewrite prompts before they reach the model and fold your conventions in automatically. For an admin-turned-developer team with inconsistent Apex and LWC habits, that is a quiet consistency engine.
4. A mod can put CI and deployment status beside the conversation.
The launch examples show custom panes rendering alongside the chat — tabs, buttons, text fields. Imagine your Apex test run or the last deployment validation visible in a pane next to the session instead of buried in terminal output.
The catch, and it is a real one: mods are not sandboxed. Anthropic's own launch post is blunt — a mod runs with the same access to your machine as Claude Code itself, so you should only install mods from sources you trust, the same way you'd install any code on your computer. For Salesforce teams this cuts both ways: the power to intercept every tool call is exactly what makes a mod useful for guardrails, and exactly why you audit anything you install. Prefer mods your team wrote or asked Claude Code to generate in-house, review the TypeScript before installing, and keep approvals on for anything that modifies your org.
Mods turn Claude Code from a tool you configure into a platform you extend. Start with one: a secret-redacting mod around Bash and Read, written with Claude Code's help in your sandbox project, tested on a harmless file, then kept on for everything after. That single mod is worth more than a page of house rules.
Originally published at Way2Force.